Abstract
The application of machine learning algorithms to predict the mechanism of action (MoA) of drugs can be highly valuable and enable the discovery of new uses for known molecules. The developed methods are usually evaluated with small subsets of MoAs with large support, leading to deceptively good generalization. However, these datasets may not accurately represent a practical use, due to the limited number of target MoAs. Accurate predictions for these rare drugs are important for drug discovery and should be a point of focus. In this work, we explore different training strategies to improve the performance of a well established deep learning model for rare drug MoA prediction. We explored transfer learning by first learning a model for common MoAs, and then using it to initialize the learning of another model for rarer MoAs. We also investigated the use of a cascaded methodology, in which results from an initial model are used as additional inputs to the model for rare MoAs. Finally, we proposed and tested an extension of Mixup data augmentation for multilabel classification. The baseline model showed an AUC of 73.2% for common MoAs and 62.4% for rarer classes. From the investigated methods, Mixup alone failed to improve the performance of a baseline classifier. Nonetheless, the other proposed methods outperformed the baseline for rare classes. Transfer Learning was preferred in predicting classes with less than 10 training samples, while the cascaded classifiers (with Mixup) showed better predictions for MoAs with more than 10 samples. However, the performance for rarer MoAs still lags behind the performance for frequent MoAs and is not sufficient for the reliable prediction of rare MoAs.

Abstract
The most malignant tumors of the central nervous system are adult-type diffuse gliomas. Historically, glioma subtype classification has been based on morphological features. However, since 2016, WHO recognizes that molecular evaluation is critical for subtyping. Among molecular markers, the mutation status of IDH1 and the codeletion of 1p/19q are crucial for the precise diagnosis of these malignancies. In pathology laboratories, however, manual screening for those markers is time-consuming and susceptible to error. To overcome these limitations, we propose a novel multimodal biomarker classification method that integrates image features derived from brain magnetic resonance imaging and histopathological exams. The proposed model consists of two branches, the first branch takes as input a multi-scale Hematoxylin and Eosin whole slide image, and the second branch uses the pre-segmented region of interest from the magnetic resonance imaging. Both branches are based on convolutional neural networks. After passing the exams by the two embedding branches, the output feature vectors are concatenated, and a multi-layer perceptron is used to classify the glioma biomarkers as a multi-class problem. In this work, several fusion strategies were studied, including a cascade model with mid-fusion; a mid-fusion model, a late fusion model, and a mid-context fusion model. The models were tested using a publicly available data set from The Cancer Genome Atlas. Our cross-validated classification models achieved an area under the curve of 0.874, 0.863, and 0.815 for the proposed multimodal, magnetic resonance imaging, and Hematoxylin and Eosin stain slide images respectively, indicating our multimodal model outperforms its unimodal counterparts and the state-of-the-art glioma biomarker classification methods.

Abstract

Abstract
Capturing high-resolution imagery of the Earth's surface often calls for a telescope of considerable size, even from low Earth orbits (LEOs). A large aperture often requires large and expensive platforms. For instance, achieving a resolution of 1 m at visible wavelengths from LEO typically requires an aperture diameter of at least 30 cm. Additionally, ensuring high revisit times often prompts the use of multiple satellites. In light of these challenges, a small, segmented, deployable CubeSat telescope was recently proposed creating the additional need of phasing the telescope's mirrors. Phasing methods on compact platforms are constrained by the limited volume and power available, excluding solutions that rely on dedicated hardware or demand substantial computational resources. Neural networks (NNs) are known for their computationally efficient inference and reduced onboard requirements. Therefore, we developed a NN-based method to measure co-phasing errors inherent to a deployable telescope. The proposed technique demonstrates its ability to detect phasing errors at the targeted performance level [typically a wavefront error (WFE) below 15 nm RMS for a visible imager operating at the diffraction limit] using a point source. The robustness of the NN method is verified in presence of high-order aberrations or noise and the results are compared against existing state-of-the-art techniques. The developed NN model ensures its feasibility and provides arealistic pathway towards achieving diffraction-limited images. (c) 2024 Optica Publishing Group

Abstract
Point source detection algorithms play a pivotal role across diverse applications, influencing fields such as astronomy, biomedical imaging, environmental monitoring, and beyond. This article reviews the algorithms used for space imaging applications from ground and space telescopes. The main difficulties in detection arise from the incomplete knowledge of the impulse function of the imaging system, which depends on the aperture, atmospheric turbulence (for ground-based telescopes), and other factors, some of which are time-dependent. Incomplete knowledge of the impulse function decreases the effectiveness of the algorithms. In recent years, deep learning techniques have been employed to mitigate this problem and have the potential to outperform more traditional approaches. The success of deep learning techniques in object detection has been observed in many fields, and recent developments can further improve the accuracy. However, deep learning methods are still in the early stages of adoption and are used less frequently than traditional approaches. In this review, we discuss the main challenges of point source detection, as well as the latest developments, covering both traditional and current deep learning methods. In addition, we present a comparison between the two approaches to better demonstrate the advantages of each methodology.